Global Monsoon in ICON: The Scale-Dependent Response of Northern Hemisphere Monsoons

Abstract. The global monsoon system is a lifeline for two-thirds of the world’s population, as it is essential for tropical water security, food, and agriculture. However, its complex multiscale interactions challenge weather and climate models. This study investigates how horizontal grid spacing (80 km, 40 km and 10 km) in the ICOsahedral Non-hydrostatic (ICON) model affects both the mean state and the variability of Northern Hemisphere monsoons across diurnal, intraseasonal, and interannual timescales. The simulations were conducted with a refactored ICON model using a Python-based dynamical core optimized for Graphics Processing Units (GPUs). All ICON simulations show substantial skill in capturing the global monsoon system domain, its onset, and its mean precipitation with a pattern correlation of > 0.7 and RMSE < 3 mm/day. For the key Northern Hemisphere regional monsoons,  South Asia (SAsiaM), West Africa (WAfriM) and North America (NAmerM), ICON achieves an accuracy > 80 % in capturing the observed monsoon domain. Crucially, the impact of grid spacing is strongly region-dependent and non-systematic. The finer grid spacing induces higher mean precipitation biases over continental SAsiaM, and WAfriM. Some of these biases are related to the intensity and location of moist monsoonal low-level jets, as well as their sensitivity to grid spacing. We further find that finer grid spacing overestimates monsoon precipitation variability at interannual and intraseasonal (high and low-frequency) scales, including intense precipitation frequency (> 10 mm/day), compared to observational references. Sensitivity tests confirm this variance amplification is a genuine model response, though it primarily reflects an overproduction of intense rainfall, while organized moderate variability may be underrepresented. This amplification stems primarily from enhanced intense grid-scale precipitation, while convective precipitation exhibits limited sensitivity to grid spacing. Process-oriented investigation show that the increased variance in the 10 km simulation over the core monsoon regions at high-frequency intraseasonal scales is linked to more intense low-pressure synoptic systems over SAsiaM and intense African easterly wave activity over WAfriM. Over NAmerM, biases are smaller and show minimal sensitivity to model grid spacing. All simulations have an excellent representation of the diurnal precipitation peak timing, with the 10 km simulation marginally performing better over continents. Our results demonstrate that increased grid spacing alone does not uniformly improve monsoon simulations. Instead, some features, such as the precipitation diurnal cycle, are improved while existing biases in mean precipitation and variability are enhanced. This underscores the role of region-dependent sensitivity of grid spacing governing monsoon dynamics.